Self-adaptive whole-body control method suitable for quadruped robot in uncertain dynamic environment

Abstract

The invention belongs to the technical field of robot control, and particularly relates to a self-adaptive whole-body control method suitable for a quadruped robot in an uncertain dynamic environment, and the method integrates an extended state observer (ESO), a model prediction controller (MPC) and a whole-body controller (WBC) based on hierarchical optimization. And a closed-loop control architecture is constructed to improve the robustness and motion coordination ability of the quadruped robot in a complex environment. Firstly, external disturbance is estimated and compensated in real time through ESO; then, the MPC optimizes future system behaviors in a rolling manner under the condition of considering interference terms, and an expected foot end counter-acting force is output; and finally, the WBC is combined with the priorities of the multiple tasks to generate all joint control instructions in a constraint optimization form, and task collaboration and dynamic adaptation are achieved. The method has the advantages of being high in disturbance self-adaptive capacity, high in control precision, suitable for a multi-task environment and the like, and is particularly suitable for quadruped robot control tasks in complex dynamic scenes such as transportation and search and rescue.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of robot control, and particularly relates to a self-adaptive whole-body control method suitable for a quadruped robot in an uncertain dynamic environment. BACKGROUND

[0002] Quadruped robots are widely used in complex scenes such as transportation, search and rescue, and emergency due to their excellent terrain adaptability. Current mainstream quadruped robots, such as ANYmal, MiniCheetah, UnitreeGo1, etc., generally use model-based control methods to achieve stable motion control.

[0003] However, in unstructured or dynamically changing environments, quadruped robots often face problems such as external disturbances, load changes, and model uncertainties, which seriously affect their trajectory tracking accuracy and stability. Although traditional robust control methods have certain anti-interference ability, they have high control frequency and strong parameter dependence, which can easily cause system oscillation. Model predictive control (MPC) is widely used in dynamic gait control due to its rolling optimization and multi-constraint processing capabilities, but its performance is highly sensitive to the accuracy of the dynamic model, making it difficult to effectively respond to unknown disturbances.

[0004] Therefore, the application proposes a self-adaptive whole-body control method suitable for a quadruped robot in an uncertain dynamic environment. The composite control strategy combining the extended state observer (ESO) and MPC improves the motion robustness of the robot and effectively tracks the centroid trajectory, body posture, and foot swing trajectory. However, in most scenarios, quadruped robots are in unstructured environments, and it is obviously extremely challenging to track all state trajectories comprehensively when the quadruped robot is moving at high speed. The quadruped robot is a floating base system with redundant degrees of freedom, which can be used to control the whole body of the robot to achieve more flexible and efficient motion. Therefore, the application proposes an adaptive whole-body control algorithm that has real-time disturbance estimation capability, can coordinate multi-task control, and is suitable for adaptive control of quadruped robots in complex dynamic environments to improve their motion performance and environmental adaptability. SUMMARY

[0005] The purpose of the application is to provide a self-adaptive whole-body control method suitable for a quadruped robot in an uncertain dynamic environment, which can be based on the fusion of an extended state observer (ESO) and a model predictive control (MPC), and can improve the robustness and motion performance of the quadruped robot in unstructured terrain by estimating and compensating external disturbances in real time. At the same time, a task priority mechanism is introduced, combined with a whole-body control (WBC) strategy, to realize the coordinated optimization between multiple control objectives.

[0006] The technical solutions adopted by the application are as follows:

[0007] An adaptive whole-body control method for quadruped robots in uncertain dynamic environments, comprising the following steps:

[0008] Step 1: Constructing a dynamic model of the quadruped robot;

[0009] Step 2: Collecting robot sensor data and using an extended state observer (ESO) to estimate system state and external disturbances in real time;

[0010] Step 3: Inputting the estimated state into a model predictive controller (MPC) with the desired state, and rolling optimization to generate desired foot-end reaction forces or accelerations;

[0011] Step 4: Using a WBC controller based on hierarchical optimization to control the task space planning tasks in a hierarchical manner;

[0012] Step 5: Using a hierarchical optimization control algorithm to convert the control target into a QP problem with equality constraints or inequality constraints, and then solving the desired robot state to achieve the desired tracking target;

[0013] Step 6: The actual execution result is fed back to the state estimator and the extended state observer ESO, forming a closed-loop control structure of "observation-prediction-optimization", and adjusting the control strategy in real time to adapt to environmental changes and disturbance effects.

[0014] The extended state observer of the method designs a multi-order position ESO and an attitude ESO respectively;

[0015] The position ESO is used to estimate the external disturbance force in three directions of the robot;

[0016] The attitude ESO is used to estimate the disturbance torque around the roll, pitch, and yaw axes;

[0017] The MPC state space of the method constructs the disturbance term estimated by the ESO as a compensation amount in the discrete state space model, achieving robustness enhancement of the controller without increasing the burden of additional sensors.

[0018] The MPC rolling optimization of the method specifically adopts a rolling horizon control strategy to solve the finite horizon optimal control problem with disturbance compensation in real time within each control period, obtaining the desired foot-end force or joint torque, and taking it as the output of the controller to act on the quadruped robot.

[0019] The method adopts a top-bottom hierarchical control architecture, in which the whole-body controller (WBC) is at the top layer, and based on the rough foot-end reaction force prediction provided by the model predictive controller (MPC), further fine control and task allocation are performed on multiple task space targets.

[0020] The upper controller is composed of an adaptive model predictive controller (MPC) and a whole-body controller (WBC) based on priority-based hierarchical optimization.

[0021] The WBC based on hierarchical optimization is a quadratic programming problem that converts tasks into a series of linear equality constraints or inequality constraints, and then solves the desired robot state, and finally substitutes it into the inverse dynamics equation to calculate the required joint torque.

[0022] The optimization strategy solves the task with high priority first, and projects the low-priority task into its null space to solve it, ensuring that the control effect of the high-priority task is not affected.

[0023] The technical effects achieved by the present application are:

[0024] The composite control strategy combining the extended state observer (ESO) and the MPC improves the motion robustness of the robot, and effectively tracks the centroid trajectory, body attitude, foot swing trajectory and the like. However, in most scenarios, the quadruped robot is in an unstructured environment, and when the quadruped robot moves at high speed, it is obviously extremely challenging to comprehensively track all state trajectories. The quadruped robot is a floating base system with redundant degrees of freedom, and these redundant degrees of freedom can be used for whole-body coordinated control of the robot, thereby realizing more flexible and efficient motion. Therefore, the present application proposes an adaptive whole-body control algorithm, which has real-time disturbance estimation capability, can coordinate multi-task control, and is suitable for an adaptive control method of a quadruped robot in a complex dynamic environment, so as to improve the motion performance and environmental adaptability of the quadruped robot.

[0025] The present application proposes a whole-body motion control scheme composed of an ESO-MPC and a WBC controller based on hierarchical optimization. The foot end force is optimized by using the ESO-MPC, and then the WBC controller based on hierarchical optimization is used for hierarchical control of the task planned in the task space, and the control target is converted into a QP problem of equality constraints or inequality constraints, and then the desired robot state is solved to achieve the desired tracking target. BRIEF DESCRIPTION OF DRAWINGS

[0026] Figure 1 is the adaptive whole-body control framework of the quadruped robot in the present application;

[0027] Figure 2 is the body height and speed tracking effect diagram of the quadruped robot in the present application under the tree root terrain;

[0028] Figure 3 is the joint position tracking effect diagram of the quadruped robot in the present application under the tree root terrain;

[0029] Figure 4is the body height and speed tracking effect diagram in the invention on the gravel road;

[0030] Figure 5 is the joint position tracking effect diagram in the invention on the gravel road;

[0031] Figure 6 is the body height and speed tracking effect diagram in the invention under the slope terrain;

[0032] Figure 7 is the joint position tracking effect diagram in the invention under the slope terrain. DETAILED DESCRIPTION

[0033] In order to make the purpose and advantages of the present application more clear and apparent, the present application will be specifically described below in conjunction with embodiments. It should be understood that the following text is only used to describe one or several specific embodiments of the present application, and does not strictly limit the specific protection scope requested by the present application.

[0034] As shown in Figures 1-7 , an adaptive whole-body control method suitable for a quadruped robot in an uncertain dynamic environment, comprising the following steps:

[0035] Step 1: constructing a dynamic model of the quadruped robot; the dynamic model of the quadruped robot includes: a rigid body dynamics model of the quadruped robot, a center of mass dynamics of the quadruped robot, and a single rigid body dynamics model;

[0036] The mathematical form of the rigid body dynamics model of the quadruped robot is represented as:

[0037]

[0038] wherein represents a generalized position vector, represents a generalized velocity vector, represents a generalized acceleration vector. is the inertia matrix of the robot system, represents a centrifugal force and Coriolis force vector, represents a gravity term. represents a selection matrix, the purpose is to convert the driving joint torque to joint torque, represents the driving joint torque, represents a conversion matrix composed of the contact Jacobian matrices corresponding to the four legs, the purpose is to convert the ground reaction force at the foot end into the leg joint torque. represents the foot end reaction force;

[0039] The general dynamics of the quadruped robot is described, which is divided into an underactuated body part and a leg-driven part. The underactuated body part (1.2) contains six degrees of freedom of the body position and attitude, and the leg-driven part of the robot (1.3) has twelve degrees of freedom, n = 12 representing the number of driven joints.

[0040]

[0041] The inverse dynamics equation is obtained from (1.4) as follows:

[0042]

[0043] Preferably, the center-of-mass dynamics of the quadruped robot

[0044] The center-of-mass dynamics of the quadruped robot is obtained on the basis of the rigid body dynamics model, extracts its underactuated part, and is converted to the center-of-mass coordinate system. The center-of-mass dynamics of the quadruped robot describes the evolution law of the overall momentum of the robot, and reveals the relationship between the momentum change and the external force through the force analysis at the center-of-mass. The center-of-mass dynamics framework of the quadruped robot provides key theoretical support for the gait planning, balance control and dynamic motion of the quadruped robot. The mathematical description of the center-of-mass dynamics of the quadruped robot is shown in (1.5).

[0045]

[0046] Its represents the center-of-mass momentum, which is composed of linear momentum and angular momentum , represents the contact force vector of each leg (i = 1, 2, 3, 4), n c represents the number of legs in contact with the ground. represents the gravity acceleration vector, represents the coordinate vector from the center-of-mass coordinate position of the robot in the world coordinate system to the foot coordinate position, represents the torque received by the robot in the environment.

[0047] Preferably, the single rigid body model is a commonly used simplified dynamics modeling method, and its core assumptions include: rigid body assumption (regarding the robot body as a single rigid body, ignoring the influence of structural flexibility); mass centralization (assuming that the total mass of the system is completely concentrated in the body, ignoring the mass distribution and inertia effect of the legs); force simplification (only considering the contact force of the supporting leg on the body, without separately modeling the leg dynamics); the form of the single rigid body dynamics model of the quadruped robot is as shown in (1.6),

[0048]

[0049] wherein mass of the robot, denotes the center of mass of the robot. The gravity vector is denoted by , corresponds to the base angular velocity of the robot. The center of mass position is denoted by The foot end positions are denoted by , where i = 1, 2, 3, 4. The contact forces of each leg are denoted by ,

[0050] Considering the influence of unknown disturbances on the robot during motion, a single rigid body dynamics model incorporating disturbance terms is established:

[0051]

[0052] The uncertainty of the robot in the translational direction is denoted by The uncertainty in the rotational direction is denoted by ; in equation (1.8) is expanded as follows:

[0053]

[0054] Step 2: Collect robot sensor data and use an extended state observer (ESO) to estimate the system state and external disturbance in real time;

[0055] Preferably, the extended state observer estimation method of step 2 specifically includes:

[0056] The present application uses a nonlinear control method of extended state observer to estimate the unknown disturbance force F b and the uncertainty torque d acting on the robot; the extended state observer includes an extended state observer based on disturbance force and an extended state observer based on disturbance torque, and the specific design method is as follows:

[0057] First, the extended state observer based on disturbance force is designed

[0058] According to equation (1.8), we have:

[0059]

[0060] where, is the control input term; is the position disturbance force acting on the center of mass of the robot;

[0061] is the n-order position ESO designed to estimate the unknown term Δ v , let denote the estimates of p, v, Δ v respectively; the n-order position ESO based on disturbance force is designed as:

[0062]

[0063] where, denote the i-th element of ; u v,i is the i-th element of u v ; is the ESO tunable parameter.

[0064] Secondly, the disturbance torque based ESO design; according to formula (1.8) and (1.9), the dynamic model can be rewritten as:

[0065]

[0066] where, is the control input; is the disturbance torque;

[0067] is designed to estimate the unknown disturbance term Δ ω , let denote the estimation of ω b , Δ ω ; the attitude ESO design based on disturbance torque is:

[0068]

[0069] where, u ω,i is the i-th element of u ω , is the attitude ESO adjustable bandwidth vector, are the i-th element of and .

[0070] Step 3: input the estimated state into the model predictive controller (MPC) with the desired state, and roll optimization to generate the desired foot end reaction force or acceleration;

[0071] Preferably, the model predictive control method of the fusion extended state observer of the step 3 specifically comprises:

[0072] The present application combines the above-mentioned n-order position ESO based on disturbance force and attitude ESO based on disturbance torque with MPC; develop an ESO-based MPC framework for compensating various uncertainties of the quadruped robot;

[0073] Based on the implementation of the quadruped robot model predictive control design of the extended state observer, after estimating the two disturbance terms, the MPC framework is integrated to ensure the friction constraint of the ground reaction force. Finally, the desired foot end force and the correction force generated by the swing leg dynamics are optimized together to be converted into joint output torque τd .

[0074] It is worth mentioning here that, in consideration of the aforementioned continuous dynamics model and the design of the extended state observer, the state space expression form of the quadruped robot system is further constructed for facilitating the design of the controller, which is used for the subsequent optimization modeling of the model predictive controller.

[0075] The discrete-time state space model of the quadruped robot system can be shown as follows:

[0076]

[0077] where the state variable the input control is the ground reaction force, is the gravity term, is the disturbance estimated by the ESO, H n ,B n ,P n is the system matrix.

[0078] Subsequently, a linear MPC problem is formulated with a horizon length of n, as follows:

[0079]

[0080] where, denotes the system state at time i, denotes the desired state of the system at the same time. F i is the ground reaction force calculated at time step i, O i and J i are diagonal positive semi-definite matrices; the components of the contact force vector of each foot are denoted as F ix , F iy and F iz . In addition, μ denotes the friction coefficient between the contact and the ground, T i is the inequality constraint matrix, S i is the force constraint matrix of the swing leg.

[0081] Step 4: Hierarchical control of the task of the task space planning by using the WBC controller based on hierarchical optimization;

[0082] Preferably, the WBC hierarchical control task of step 4 specifically includes:

[0083] Next, each task will be described in detail, and the WBC task priority and type are shown in Table 1.

[0084]

[0085]

[0086] Table 1: WBC task priorities and types

[0087] Dynamic consistency tasks:

[0088] The dynamic consistency task describes the robot's movement in the process of satisfying the generalized acceleration defined in the whole body dynamics equation Foot contact force F c The relationship between and joint torque τ; the dynamic equation is as follows:

[0089]

[0090] Torque limiting task:

[0091] The torque limit task describes the limitation measures on the maximum torque output of the joint motor to prevent damage to the motor during the robot's movement. The torque limiting task is:

[0092]

[0093] in

[0094] Contact point motion constraint task

[0095] The contact point motion constraint task describes the point contact when the supporting leg touches the ground during the robot's movement. In order to reduce the impact of the ground reaction force on the robot, the speed and acceleration of the supporting leg foot end are constrained to zero to ensure that the robot stably contacts the ground during the movement.

[0096]

[0097] The mathematical description of this task can be written as:

[0098]

[0099] Friction cone constraint task

[0100] The friction cone is a mathematical model used to describe the friction force limit range of the contact point. The friction cone constraint task describes the robot's movement to prevent the supporting leg from slipping or standing unstable when it contacts the ground. Therefore, the foot contact force 3 needs to be restricted within the friction cone, and the contact force in the z direction is greater than zero; in order to simplify the friction cone constraint, the support leg is constrained by the friction tetrahedron to obtain a linear friction constraint; the support leg contact force f i =[f i x,f i y,f i z]T, so the friction pyramid constraint of the i-th supporting leg is:

[0101] μ f f i ≤0 5×1 ,(1.21)

[0102] where μ denotes the friction coefficient, μ f represents the frictional four-pyramid matrix,

[0103]

[0104] The friction cone mission inequality constraint can be written as:

[0105] [0 10×18 S cμ (μ f ) 0 10×12 ]u d ≤0 10×1 , (1.23)

[0106] where denotes the friction selection constraint matrix;

[0107] Swing leg force constraint mission

[0108] The swing leg force constraint mission describes that during the robot motion process, the robot ensures that the swing leg does not exert additional force to the ground during the swing phase, and the foot end force constraint is zero, avoiding the swing trajectory from deviating from the expected path due to external force. The force constraint matrix can be expressed as:

[0109]

[0110] where denotes the swing leg selection matrix;

[0111] Swing leg trajectory tracking mission

[0112] The swing leg trajectory tracking mission describes that during the robot motion process, the swing leg ensures that it can accurately move according to the planned trajectory during the swing phase, and smoothly land within the specified time, while the swing leg foot end acceleration tracks the expected foot end acceleration to achieve the desired motion of the body; the single leg foot end acceleration can be described as:

[0113]

[0114] The expected foot end acceleration is as follows:

[0115]

[0116] where denote the foot end position error and velocity error, respectively, where denotes the corresponding proportional-derivative coefficient; up to this point the task is expressed as:

[0117]

[0118] Therefore, the task is written as a unified linear equality constraint problem:

[0119]

[0120] where denotes the Jacobian matrix of the swing leg;

[0121] The body acceleration tracking task:

[0122] The body acceleration tracking task describes that during the robot motion, the body should satisfy the planned linear acceleration and angular acceleration of the center of mass. The body acceleration tracking task is written as a unified equality constraint:

[0123]

[0124] The contact force tracking task:

[0125] The contact force tracking task describes that during the robot motion, the support leg contact force satisfies the foot end force fmpc trajectory curve solved by the model predictive controller, and this task has the lowest priority; the contact force tracking task is expressed as a unified linear equality constraint:

[0126]

[0127] where denotes the contact Jacobian matrix;

[0128] Step 5: A hierarchical optimization control algorithm is used to convert the control target into a QP problem with equality constraints or inequality constraints, and then the expected robot state is solved to achieve the expected tracking target;

[0129] Preferably, the hierarchical optimization control algorithm of step 5 specifically includes:

[0130] The task T is described as a solution vector a set of linear equations or inequalities above:

[0131]

[0132] where u and v are relaxation variables to be minimized, and for T1, T2...T w W tasks in strict priority order, finally obtaining an optimal solution x * ; In order to ensure the strict task priority order, it can be selected that all higher priority equality constraints Z w=N( A w ) to find the solution x for the next task in the null space w+1 ,in Finally, we get the solution vector x w+1 =x * +Z w z w+1 , where z w+1 In Z w Vector in row space. Solve task T w+1 It is equivalent to solving the following QR problem and

[0133]

[0134] Set variable Thus, the QR problem described in (1.32) can be written as:

[0135]

[0136] in

[0137]

[0138] To speed up computation time, an orthogonal triangular decomposition is used to implement the equality constraints. A The null space basis of Z ω Decomposition; but as more equations are added to the task stack, the calculation of the null space of A will become slower, so an iterative method is used to speed up the calculation; given two full row rank matrices A1 and A2, Nullspace basis of Z2

[0139] Z2=N(A1)N(A2N(A1))=Z1N(A2Z1),(1.34)

[0140] It can be shown that equation (1.35) is the nullspace basis of A1 and A2:

[0141]

[0142] Step 6: The actual execution results are fed back to the state estimator and the extended state observer (ESO), forming a closed-loop control structure of "observation-prediction-optimization" to adjust the control strategy in real time to adapt to environmental changes and disturbances.

[0143] Implementation methods of the experimental verification and analysis of the present invention:

[0144] To verify the effectiveness of the adaptive whole-body control method for quadruped robots in uncertain dynamic environments proposed in the present application, a load transportation test on various complex terrains is designed to verify the effectiveness of the proposed adaptive whole-body motion control scheme. These test terrains include forest terrain with tree roots, slope terrain, and gravel road surface. Through experiments in these complex environments, the stability, adaptability and motion control performance of the robot when facing unknown loads and complex terrains are evaluated. The real machine test platform is the Go1 quadruped robot produced by Unitree Robotics Company.

[0145] Scenario One: Forest Terrain Transportation Test

[0146] Through the test, the quadruped robot successfully traverses the forest terrain with tree roots and the disturbance of unmodeled load to verify the robustness and adaptability of the proposed control method in unstructured environments. In the experimental test, the Go1 robot carries heavy objects and walks on the terrain, and successfully passes through the terrain to complete the load transportation task. This experimental process shows that the designed control method can effectively cope with external disturbances in complex terrain and ensure the reliability and task execution ability of the robot in unstructured environments.

[0147] As Figure 2 shown, the figure shows the changes in body height and speed of the Go1 robot during the transportation task in the forest terrain. As shown in the data in the figure, the robot maintains good body tracking ability during the transportation process. Even in the case of instantaneous speed increase caused by the foot touching the tree roots, the robot can adjust itself in a short time and restore stability. This shows that the robot has good environmental adaptability and anti-interference ability.

[0148] To demonstrate and verify the effectiveness of the proposed control framework, joint position tracking data is recorded. Considering the consistency of the joint control task in the algorithm, the joint data of the right front leg is taken as a representative to demonstrate, as shown in Figure 3 , the results show that even in the forest terrain with tree roots, the robot can still achieve accurate joint position tracking, showing the superiority of the control framework in multi-task coordination and complex environment adaptation.

[0149] Scenario Two: Gravel Ground Transportation Test

[0150] Considering that the interference caused by the forest terrain is not persistent enough and the terrain is not smooth enough, the study designs and carries out the load transportation test of the quadruped robot on the gravel terrain. In the experiment, a gravel road composed of pebbles is laid, and Go1 robot also carries 4.5 kg of load to walk on the gravel terrain. The test results show that even on the smooth and continuously disturbed ground, the robot can still maintain good anti-interference ability and walking stability, successfully pass through the complex terrain and complete the load transportation task.

[0151] As Figure 4 shown, the figure shows the change of the body height and the change of the speed of Go1 robot when carrying out the transportation task on the gravel road. As can be seen from the data in the figure, throughout the transportation process, the robot always maintains good body tracking ability and speed tracking ability. Even in the face of continuous disturbance, the robot can still maintain good anti-interference ability.

[0152] Similarly, the joint data of the right front leg are taken as representatives to show the effectiveness of the proposed control framework and the actual effect of WBC multi-task control. As Figure 5 shown, the experimental results show that even on the continuously disturbed gravel ground, the robot can accurately track the joint position and exhibit the ability to quickly respond to external disturbances, further proving the adaptability and stability of the designed control framework in complex terrain environment.

[0153] Scenario three: slope terrain transportation test

[0154] Based on the characteristics of the relatively flat slope of the previous two terrains, in order to further verify the adaptability and robustness of the proposed control method in the terrain with slope, the load transportation test of the quadruped robot on the slope terrain is designed and carried out. In the experiment, Go1 robot also carries 4.5 kg of load to walk on the slope terrain. The test results show that even in the slope environment, the robot can still maintain good anti-interference ability and walking stability, successfully pass through the complex terrain and complete the load transportation task.

[0155] As Figure 6 shown, the figure shows the change of the body height and the change of the speed of Go1 robot when carrying out the transportation task on the slope terrain. As can be seen from the data in the figure, throughout the transportation process, the robot always maintains good body tracking ability and speed tracking ability. This shows that the proposed control method enables the robot to stably and efficiently cope with complex terrain with slope, further verifying the reliability of the proposed method in complex environment.

[0156] Still, the joint data of the right front leg are taken as representatives for analysis and display. As Figure 7As shown, the experimental results show that the robot can accurately track the joint position even on the slope terrain with a certain slope, and exhibits fast response to external disturbance and good robustness. This fully demonstrates that the control framework can effectively coordinate the multi-task control requirements and maintain stable motion performance in complex terrain environment, providing strong support for application in complex task scenarios.

[0157] The above only describes the preferred embodiments of the present application, and it should be noted that for those skilled in the art, without departing from the principles of the present application, a number of improvements and refinements can be made, which should also be considered as the protection scope of the present application. The structures, devices and operation methods not specifically described and explained in the present application are implemented according to the conventional means in the art, unless otherwise specified and limited.

Claims

1. An adaptive whole-body control method for a quadruped robot in an uncertain dynamic environment, characterized by: The following steps are involved: Step 1: Construct a dynamic model of the quadruped robot; Step 2: Collect robot sensor data and use the extended state observer (ESO) to estimate the system state and external disturbances in real time; Step 3: Input the estimated state and the desired state into the model predictive controller (MPC), and generate the desired foot-end reaction force or acceleration through rolling optimization. Step 4: Use the WBC controller based on hierarchical optimization to perform hierarchical control on the task space planning task; Step 5: Use a hierarchical optimization control algorithm to transform the control objective into a QP problem with equality or inequality constraints, and then solve the desired robot state to achieve the desired tracking target; Step 6: The actual execution results are fed back to the state estimator and the extended state observer (ESO), forming a closed-loop control structure of "observation-prediction-optimization". The control strategy is adjusted in real time to adapt to environmental changes and disturbances.

2. The adaptive whole-body control method for a quadruped robot in an uncertain dynamic environment according to claim 1, characterized in that: In step 1, the dynamic model of the quadruped robot includes: a quadruped robot rigid body dynamic model, a quadruped robot center of mass dynamics, and a single rigid body dynamic model.

3. The adaptive whole-body control method for a quadruped robot in an uncertain dynamic environment according to claim 2, characterized in that: The mathematical form of the quadruped robot rigid body dynamics model is expressed as: in represents the generalized position vector, represents the generalized velocity vector, represents the generalized acceleration vector; is the inertia matrix of the robot system, represents the centrifugal force and Coriolis force vector, represents the gravity term; represents the selection matrix, the purpose of which is to convert the driving joint torque to the joint torque, represents the driving joint torque, The transformation matrix composed of the contact Jacobian matrices corresponding to the four legs is used to convert the ground reaction force at the foot end into the leg joint torque; represents the reaction force at the foot end; This paper describes the general dynamics of a quadruped robot. The dynamics are divided into the underactuated part of the body and the actuated part of the legs. The underactuated part of the body (1.2) contains six degrees of freedom in terms of body position and attitude, while the actuated part of the legs has twelve degrees of freedom (1.3), where n = 12 represents the number of actuated joints. From (1.4), the inverse dynamics equation is:

4. The adaptive whole-body control method for a quadruped robot in an uncertain dynamic environment according to claim 3, characterized in that: Center of mass dynamics of the quadruped robot The center-of-mass dynamics of a quadruped robot is derived by extracting the underactuated portion of the rigid-body dynamics model and converting it to a center-of-mass coordinate system. The center-of-mass dynamics of a quadruped robot describes the evolution of the robot's overall momentum and, through force analysis at the center of mass, reveals the relationship between momentum changes and external forces. The center-of-mass dynamics framework of a quadruped robot provides key theoretical support for gait planning, balance control, and dynamic motion. The mathematical description of the center-of-mass dynamics of a quadruped robot is shown in (1.5).

5. The adaptive whole-body control method for a quadruped robot in an uncertain dynamic environment according to claim 4, characterized in that: The single rigid body dynamics model of the quadruped robot is as shown in formula (1.6): in Represents the mass of the robot, represents the moment of inertia of the robot's center of mass; the gravity vector is represented by express, The corresponding angular velocity of the robot base; the center of mass position is Foot end position Indicated by, where i = 1, 2, 3, 4; the contact force of each leg is expressed as express, Establish a single rigid body dynamics model that incorporates interference terms: The uncertainty of the robot in the translation direction is expressed as The uncertainty in the direction of rotation is expressed as In formula (1.8) Expands to the following form:

6. The adaptive whole-body control method for a quadruped robot in an uncertain dynamic environment according to claim 5, characterized in that: The extended state observer estimation method in step 2 specifically includes: The nonlinear control method of the extended state observer is used to estimate the unknown disturbance force F acting on the robot. b and uncertainty torque d; the extended state observer includes an extended state observer based on interference force and an extended state observer based on interference torque, and the specific method is as follows: First, the extended state observer is designed based on the disturbance force; according to formula (1.8), we can get in, is the control input; is the position disturbance force acting on the robot’s center of mass; Estimate the unknown term Δ for designing n-th order position ESO v ,set up Represent p, v, Δ respectively v The n-th order position ESO design based on the interference force is: in, Respectively The i-th element of v,i for u v The i-th element of ; It is an ESO adjustable parameter; Secondly: Design of extended state observer based on disturbance torque; According to formulas (1.8) and (1.9), the dynamic model can be rewritten as: in, is the control input; is the interference torque; To design an attitude ESO, estimate the unknown disturbance term Δ ω ,make Respectively represent ω b , Δ ω The attitude ESO design based on the disturbance torque is: Among them, u ω,i is u ω The i-th element of is the bandwidth vector that the attitude ESO can adjust, They are and The i-th element of .

7. The adaptive whole-body control method for a quadruped robot in an uncertain dynamic environment according to claim 6, characterized in that: The model predictive control method integrating the extended state observer in step 3 specifically includes: Combine the previously mentioned nth-order position ESO based on disturbance force and the attitude ESO based on disturbance torque with MPC; develop an ESO-based MPC framework to compensate for various uncertainties of the quadruped robot; The discrete-time state space model of the quadruped robot system can be expressed as: The state variables Input Control is the ground reaction force, is the gravity term, is the interference estimated by ESO, H n ,B n ,P n is the system matrix; Subsequently, a linear MPC problem is formulated with a horizon length of n as follows: in, represents the system state at time i, Indicates the expected state of the system at the same time; F i is the ground reaction force calculated at time step i, O i and J i is a diagonal positive semidefinite matrix; the components of the contact force vector for each foot are expressed as F ix , F iy and F iz ; In addition, μ represents the friction coefficient between the contact and the ground, T i is the inequality constraint matrix, S i is the force constraint matrix of the swing leg.

8. The adaptive whole-body control method for a quadruped robot in an uncertain dynamic environment according to claim 7, characterized in that: The tasks of the WBC layered control in step 4 specifically include: Dynamic consistency tasks: The dynamic consistency task describes the robot's movement in the process of satisfying the generalized acceleration defined in the whole body dynamics equation Foot contact force F c The relationship between and joint torque τ; the dynamic equation is as follows: Torque limiting task: make The torque limiting task is: in The contact point motion constraint task is: The mathematical description of this task can be written as: Friction Cone Constraint Task: The support leg is constrained by using a friction pyramid to obtain a linear friction constraint; the support leg contact force f i =[f ix ,f iy ,f iz ]T, so the friction pyramid constraint of the i-th supporting leg is: m f f i ≤0 5×1 ,(1.21) Where μ represents the friction coefficient, μ f represents the friction tetrahedron matrix, The friction cone task inequality constraint is written as: [0 10×18 S cμ (μ f )0 10×12 ]u d ≤0 10×1 ,(1.23) in represents the friction selection constraint matrix; Swing leg force constraint task: The constraint matrix is ​​expressed as: in represents the swing leg selection matrix; Swing leg trajectory tracking task: The acceleration of the foot end of a single leg can be described as: Expected foot acceleration As shown below: in represent the foot end position error and velocity error respectively, Represents the corresponding proportional differential coefficient; so far the task is expressed as: Therefore, the task is written as a unified linear equality constraint problem: in The Jacobian matrix representing the swing leg composition; Airframe acceleration tracking task: The airframe acceleration tracking task is written in a unified equality constraint form: Contact force tracking task: The contact force tracking task is expressed as a unified linear equality constraint: in represents the contact Jacobian matrix.

9. The adaptive whole-body control method for a quadruped robot in an uncertain dynamic environment according to claim 8, characterized in that: The hierarchical optimization control algorithm of step 5 specifically includes: The task T is described as the solution vector A set of linear equality or inequality constraints on : Where u and v are the slack variables to be minimized, for T1, T2...T w Wait for W tasks, solve them in strict priority order, and finally get an optimal solution x * ; Select all higher priority equality constraints Z w =N( A w ) to find the solution x for the next task in the null space w+1 ,in Finally, we get the solution vector x w+1 =x * +Z w z w+1 , where z w+1 In Z w Vector in row space; solve task T w+1 It is equivalent to solving the following QR problem and Set variable Thus, the QR problem described in (1.32) can be written as: in Use orthogonal triangular decomposition to implement the null space basis Z of the equality constraint A ω Decomposition; Use iterative methods to speed up the calculation; Given two full row rank matrices A1 and A2, The null space basis of Z2: Z2=N(A1)N(A2N(A1))=Z1N(A2Z1),(1.34) Prove that equation (1.35) is the nullspace basis of A1 and A2:

Images